Private AI deployment
Self-hosted AI on your own server, set up properly
Self-hosted AI means your app and model run on a server you control, so your documents and conversations never pass through someone else’s service. I deploy AI apps and private chatbots on your own VPS or AWS account in fixed packages, with the parts most deployments skip: security, backups, monitoring and a way to roll back.
Fixed packages from $300 · I confirm which package fits before you pay

Three fixed packages
Single App Deploy
$300
One app on your VPS or AWS EC2: Docker, nginx, your domain, SSL and server hardening, with a 30-minute handover.
App + Database + CI/CD
$750
Everything in Single App Deploy, plus a database, CI/CD that redeploys on every push, logs and uptime alerts, with a 45-minute handover.
Full AI Stack on VPS or AWS
$1,500
Up to three services, load balancing, IAM and VPC, backups and alerts, a one-hour handover and 30 days of support.
Larger jobs start with the $150 scoping session and a fixed quote at $80 to $100 an hour. Server, domain and model costs are billed to you directly.
Why “it works on my laptop” isn’t a deployment
Getting an AI app running on a server takes an afternoon. Keeping it running is the real job: SSL that renews, a firewall that’s actually closed, backups that restore, updates that don’t take the site down, and alerts before your users notice a problem. Those are the parts I set up and hand over in writing.
What gets set up, and what you receive
Every package leaves your server in a state someone else could take over: the app running in Docker, nginx in front with SSL that renews by itself, a firewall that only opens what’s needed, and secrets kept out of the code. The larger packages add a database, CI/CD so a push to your repository redeploys the app, logs, uptime alerts and nightly backups that restore with one command. You get a written handover note covering where everything runs, how to update it, how to roll back and who to call, plus the handover call itself.
Running a model on your own server
If your documents can’t leave your infrastructure, the model can run on the same server with Ollama or a similar runtime. That removes per-message API fees and keeps every conversation private, at the cost of needing a server with enough memory, and sometimes a GPU, for the model you choose. I’ll tell you what your use case needs before you pay, and if a hosted model through a private API is the better trade-off, I’ll say so.
Proof: Bastion, a private document chatbot on your own server
Bastion answers questions from PDFs, Word files and web pages and cites the passage, page and section behind every answer. The language model runs locally with Ollama, so no data leaves the server and there are no per-message API fees. One install script sets up Docker, the firewall, nginx and SSL on an Ubuntu VPS or AWS EC2; it has user roles, rate limits and an audit log, nightly backups that restore with one command, and 83 automated tests. Built with Python, FastAPI, Qdrant and Ollama.
For clients, I’ve deployed a startup’s product on a VPS (the founder came back for a second one) and moved a company website from shared hosting onto a VPS alongside a running production app without disrupting it.
Who it’s for
Startups and small companies with a working AI app that needs a proper home, teams whose data has to stay on their own infrastructure, and founders who want to own their server rather than rent a platform. If the app doesn’t run yet, start with a build instead.
Questions
Who pays for the server and the model?
You do, directly, so you keep full access and see every cost.
VPS or AWS?
A VPS is simpler and usually cheaper for one or two services; AWS suits you if you already use it or need its network and access controls. I’ll recommend one before you pay.
Can the model run locally?
Yes, with Ollama or a similar runtime, if your server has the resources. I’ll tell you what that needs.
What happens after delivery?
30 days of fixes to anything delivered on every package. Restarts, config changes and new requests are included for 30 days in Full AI Stack and cost $250 for 30 days on the other two. After that, a monthly plan from $250.
Can you deploy an app I didn’t build with you?
Yes, as long as it runs on your own computer first. Fixes to the app itself are quoted separately.
Do I need to know Linux?
No. The handover note is written for someone who doesn’t, and the monthly plan covers anything you’d rather not touch.
Tell me what you want to deploy
From $300: one app deployed in 3 days.